A traditional randomized clinical trial is the established gold standard for estimating the causal effects of treatments. However, for some treatments, such a randomized trial cannot be performed due to ethical and other reasons. One way around this impasse is to perform a randomized encouragement design study (EDS). In the randomized EDS, clinicians are randomly assigned to receive or not to receive an encouragement for the use of the treatment on their patients. Because the randomization to encouragement leads to a natural instrumental variable under some plausible assumptions, the randomized EDS provides a tool for estimating causal effects of the treatment on patient outcomes. However, rigorous evaluation about the causal treatment effect in a EDS is often difficult because of the issue of clinician non-compliance with the intervention in EDS. In this proposal, we will develop a general statistical methodology for causal treatment analysis of the randomized EDS when data have a complicated structure.